SlideShare uma empresa Scribd logo
1 de 73
Baixar para ler offline
Lecture 7
Artificial neural networks:
Supervised learning
 Introduction, or how the brain works
 The neuron as a simple computing element
 The perceptron
 Multilayer neural networks
 Accelerated learning in multilayer neural networks
 The Hopfield network
 Bidirectional associative memories (BAM)
 Summary
© Negnevitsky, Pearson Education, 2005

1
Introduction, or how the brain works
Machine learning involves adaptive mechanisms
that enable computers to learn from experience,
learn by example and learn by analogy. Learning
capabilities can improve the performance of an
intelligent system over time. The most popular
approaches to machine learning are artificial
neural networks and genetic algorithms. This
lecture is dedicated to neural networks.

© Negnevitsky, Pearson Education, 2005

2
 A neural network can be defined as a model of
reasoning based on the human brain. The brain
consists of a densely interconnected set of nerve
cells, or basic information-processing units, called
neurons.
 The human brain incorporates nearly 10 billion
neurons and 60 trillion connections, synapses,
between them. By using multiple neurons
simultaneously, the brain can perform its functions
much faster than the fastest computers in existence
today.
© Negnevitsky, Pearson Education, 2005

3
 Each neuron has a very simple structure, but an
army of such elements constitutes a tremendous
processing power.
 A neuron consists of a cell body, soma, a number
of fibers called dendrites, and a single long fiber
called the axon.

© Negnevitsky, Pearson Education, 2005

4
Biological neural network
Synapse
Axon

Soma
Dendrites

Synapse

Dendrites
Axon

Soma
Synapse

© Negnevitsky, Pearson Education, 2005

5
 Our brain can be considered as a highly complex,
non-linear and parallel information-processing
system.
 Information is stored and processed in a neural
network simultaneously throughout the whole
network, rather than at specific locations. In other
words, in neural networks, both data and its
processing are global rather than local.
 Learning is a fundamental and essential
characteristic of biological neural networks. The
ease with which they can learn led to attempts to
emulate a biological neural network in a computer.
© Negnevitsky, Pearson Education, 2005

6
 An artificial neural network consists of a number of
very simple processors, also called neurons, which
are analogous to the biological neurons in the brain.
 The neurons are connected by weighted links
passing signals from one neuron to another.
 The output signal is transmitted through the
neuron’s outgoing connection. The outgoing
connection splits into a number of branches that
transmit the same signal. The outgoing branches
terminate at the incoming connections of other
neurons in the network.
© Negnevitsky, Pearson Education, 2005

7
Input Signals

Output Signals

Architecture of a typical artificial neural network

Middle Layer
Input Layer

© Negnevitsky, Pearson Education, 2005

Output Layer

8
Analogy between biological and
artificial neural networks
Biological Neural Network
Soma
Dendrite
Axon
Synapse

© Negnevitsky, Pearson Education, 2005

Artificial Neural Network
Neuron
Input
Output
Weight

9
The neuron as a simple computing element
Diagram of a neuron
Input Signals
x1
x2

xn

Weights

Output Signals
Y

w1
w2

wn

© Negnevitsky, Pearson Education, 2005

Neuron

Y

Y
Y

10
 The neuron computes the weighted sum of the input
signals and compares the result with a threshold
value, θ. If the net input is less than the threshold,
the neuron output is –1. But if the net input is
greater than or equal to the threshold, the neuron
becomes activated and its output attains a value +1.
 The neuron uses the following transfer or activation
function:
n

X = ∑ xi wi
i =1

+1, if X ≥ θ
Y =
−1, if X < θ

 This type of activation function is called a sign
function.
© Negnevitsky, Pearson Education, 2005

11
Activation functions of a neuron
Step function

Sign function Sigmoid function Linear function

Y

Y

Y

Y

+1

+1

1

1

0

X

0

X

-1

-1

0
-1

1
,
step=  , if X ≥ 0 Y sign = +1 if X ≥ 0 Y sigmoid=
Y


0, if X < 0

−1, if X < 0

© Negnevitsky, Pearson Education, 2005

X

0

X

-1

1
1 + e− X

Y linear= X

12
Can a single neuron learn a task?
 In 1958, Frank Rosenblatt introduced a training
algorithm that provided the first procedure for
training a simple ANN: a perceptron.
 The perceptron is the simplest form of a neural
network. It consists of a single neuron with
adjustable synaptic weights and a hard limiter.

© Negnevitsky, Pearson Education, 2005

13
Single-layer two-input perceptron
Inputs
x1
w1

Linear
Combiner

Hard
Limiter

Output
Y

w2

x2

© Negnevitsky, Pearson Education, 2005

θ
Threshold
14
The Perceptron
 The operation of Rosenblatt’s perceptron is based
on the McCulloch and Pitts neuron model. The
model consists of a linear combiner followed by a
hard limiter.
 The weighted sum of the inputs is applied to the
hard limiter, which produces an output equal to +1
if its input is positive and −1 if it is negative.

© Negnevitsky, Pearson Education, 2005

15
 The aim of the perceptron is to classify inputs,
x1, x2, . . ., xn, into one of two classes, say
A1 and A2.
 In the case of an elementary perceptron, the ndimensional space is divided by a hyperplane into
two decision regions. The hyperplane is defined by
the linearly separable function:
n

∑ xi wi − θ = 0
i =1

© Negnevitsky, Pearson Education, 2005

16
Linear separability in the perceptrons
x2

x2
Class A 1
1
1

Class A 2

2

x1

x1

2
x 1w 1 + x 2w 2 −θ = 0
(a) Two-input perceptron.

© Negnevitsky, Pearson Education, 2005

x3

x1w 1 + x2w 2 + x3w 3 −θ = 0
(b) Three-input perceptron.

17
How does the perceptron learn its classification
tasks?
This is done by making small adjustments in the
weights to reduce the difference between the actual
and desired outputs of the perceptron. The initial
weights are randomly assigned, usually in the range
[−0.5, 0.5], and then updated to obtain the output
consistent with the training examples.

© Negnevitsky, Pearson Education, 2005

18
 If at iteration p, the actual output is Y(p) and the
desired output is Yd (p), then the error is given by:

e( p) = Yd ( p) − Y ( p)

where p = 1, 2,
3, . . .

Iteration p here refers to the pth training example
presented to the perceptron.
 If the error, e(p), is positive, we need to increase
perceptron output Y(p), but if it is negative, we
need to decrease Y(p).
© Negnevitsky, Pearson Education, 2005

19
The perceptron learning rule
wi ( p + 1) = wi ( p ) +  . xi ( p ) . e( p )
×
where p = 1, 2, 3, . . .
α is the learning rate, a positive constant less than
unity.
The perceptron learning rule was first proposed by
Rosenblatt in 1960. Using this rule we can derive
the perceptron training algorithm for classification
tasks.

© Negnevitsky, Pearson Education, 2005

20
Perceptron’s training algorithm
Step 1: Initialisation
Set initial weights w1, w2,…, wn and threshold
θ
to random numbers in the range [−0.5, 0.5].
If the error, e(p), is positive, we need to increase
perceptron output Y(p), but if it is negative, we
need to decrease Y(p).

© Negnevitsky, Pearson Education, 2005

21
Perceptron’s training algorithm (continued)
Step 2: Activation
Activate the perceptron by applying inputs x1(p),
x2(p),…, xn(p) and desired output Yd (p). Calculate
the actual output at iteration p = 1
 n

Y ( p ) = step  ∑ x i ( p ) w i ( p ) − θ 
 i =1




where n is the number of the perceptron inputs,
and step is a step activation function.

© Negnevitsky, Pearson Education, 2005

22
Perceptron’s training algorithm (continued)
Step 3: Weight training
Update the weights of the perceptron
wi ( p + 1) = wi ( p) + ∆wi ( p)
where ∆wi(p) is the weight correction at iteration p.
The weight correction is computed by the delta
rule:
.
∆wi ( p) = α ×xi ( p ) ×e( p)

Step 4: Iteration
Increase iteration p by one, go back to Step 2 and
repeat the process until convergence.
© Negnevitsky, Pearson Education, 2005

23
Example of perceptron learning: the logical operation AND
Epoch

Inputs

Desired
output
Yd

Initial
weights
w1
w2

Actual
output
Y

Error

Final
weights
w1
w2

x1

x2

1

0
0
1
1

0
1
0
1

0
0
0
1

0.3
0.3
0.3
0.2

− 0.1
− 0.1
− 0.1
− 0.1

0
0
1
0

0
0
−1
1

0.3
0.3
0.2
0.3

− 0.1
− 0.1
− 0.1
0.0

2

0
0
1
1

0
1
0
1

0
0
0
1

0.3
0.3
0.3
0.2

0.0
0.0
0.0
0.0

0
0
1
1

0
0
−1
0

0.3
0.3
0.2
0.2

0.0
0.0
0.0
0.0

3

0
0
1
1

0
1
0
1

0
0
0
1

0.2
0.2
0.2
0.1

0.0
0.0
0.0
0.0

0
0
1
0

0
0
−1
1

0.2
0.2
0.1
0.2

0.0
0.0
0.0
0.1

4

0
0
1
1

0
1
0
1

0
0
0
1

0.2
0.2
0.2
0.1

0.1
0.1
0.1
0.1

0
0
1
1

0
0
−1
0

0.2
0.2
0.1
0.1

0.1
0.1
0.1
0.1

5

0
0
1
1

0
1
0
1

0
0
0
1

0.1
0.1
0.1
0.1

0.1
0.1
0.1
0.1

0
0
0
1

0
0
0
0

0.1
0.1
0.1
0.1

0.1
0.1
0.1
0.1

Threshold: θ = 0.2; learning rate:
© Negnevitsky, Pearson Education, 2005

e

= 0.1
24
Two-dimensional plots of basic logical operations
x2

x2

x2

1

1

1
x1

x1
0

1

(a) AND (x1 ∩ x2)

0

1

(b) OR (x 1 ∪ x 2 )

x1
0

1

(c) Ex cl us iv e- OR
(x 1 ⊕ x 2 )

A perceptron can learn the operations AND and OR,
but not Exclusive-OR.
© Negnevitsky, Pearson Education, 2005

25
Multilayer neural networks
 A multilayer perceptron is a feedforward neural
network with one or more hidden layers.
 The network consists of an input layer of source
neurons, at least one middle or hidden layer of
computational neurons, and an output layer of
computational neurons.
 The input signals are propagated in a forward
direction on a layer-by-layer basis.

© Negnevitsky, Pearson Education, 2005

26
Input Signals

Output Signals

Multilayer perceptron with two hidden layers

Input
layer

First
hidden
layer

© Negnevitsky, Pearson Education, 2005

Second
hidden
layer

Output
layer

27
What does the middle layer hide?
 A hidden layer “hides” its desired output.
Neurons in the hidden layer cannot be observed
through the input/output behaviour of the network.
There is no obvious way to know what the desired
output of the hidden layer should be.
 Commercial ANNs incorporate three and
sometimes four layers, including one or two
hidden layers. Each layer can contain from 10 to
1000 neurons. Experimental neural networks may
have five or even six layers, including three or four
hidden layers, and utilise millions of neurons.
© Negnevitsky, Pearson Education, 2005

28
Back-propagation neural network
 Learning in a multilayer network proceeds the
same way as for a perceptron.
 A training set of input patterns is presented to the
network.
 The network computes its output pattern, and if
there is an error − or in other words a difference
between actual and desired output patterns − the
weights are adjusted to reduce this error.

© Negnevitsky, Pearson Education, 2005

29
 In a back-propagation neural network, the learning
algorithm has two phases.
 First, a training input pattern is presented to the
network input layer. The network propagates the
input pattern from layer to layer until the output
pattern is generated by the output layer.
 If this pattern is different from the desired output,
an error is calculated and then propagated
backwards through the network from the output
layer to the input layer. The weights are modified
as the error is propagated.
© Negnevitsky, Pearson Education, 2005

30
Three-layer back-propagation neural network
Input signals
x1
x2

xi

1
1

2

y2

k

yk

l

yl

1

2

2
i

y1

wij

j

wjk

m

xn

n

Input
layer

Hidden
layer

Output
layer

Error signals
© Negnevitsky, Pearson Education, 2005

31
The back-propagation training algorithm
Step 1: Initialisation
Set all the weights and threshold levels of the
network to random numbers uniformly
distributed inside a small range:
 2.4
2.4 
−
 F , + F ÷
÷
i
i 


where Fi is the total number of inputs of neuron i
in the network. The weight initialisation is done
on a neuron-by-neuron basis.
© Negnevitsky, Pearson Education, 2005

32
Step 2: Activation
Activate the back-propagation neural
network by applying inputs x1(p), x2(p),…, xn(p)
and desired outputs yd,1(p), yd,2(p),…, yd,n(p).
(a) Calculate the actual outputs of the neurons in
the hidden layer:
 n

y j ( p ) = sigmoid  ∑ xi ( p ) ×wij ( p ) − θ j 
 i =1




where n is the number of inputs of neuron j in the
hidden layer, and sigmoid is the sigmoid activation
function.
© Negnevitsky, Pearson Education, 2005

33
Step 2 : Activation (continued)
(b) Calculate the actual outputs of the neurons in
the output layer:
m

y k ( p ) = sigmoid  ∑ x jk ( p ) ×w jk ( p ) − θ k 
 j =1




where m is the number of inputs of neuron k in the
output layer.

© Negnevitsky, Pearson Education, 2005

34
Step 3: Weight training
Update the weights in the back-propagation
network propagating backward the errors associated
with output neurons.
(a) Calculate the error gradient for the neurons in
the output layer:
k ( p)

= yk ( p) ×1 − y k ( p ) ×ek ( p )

where ek ( p ) = yd ,k ( p ) − yk ( p )
Calculate the weight corrections:
∆w jk ( p) = ×y j ( p) × k ( p)

Update the weights at the output neurons:
w jk ( p + 1) = w jk ( p ) + ∆w jk ( p )
© Negnevitsky, Pearson Education, 2005

35
Step 3: Weight training (continued)
(b) Calculate the error gradient for the neurons in
the hidden layer:
l

j ( p)

= y j ( p ) × 1 − y j ( p )] ×∑ k ( p ) w jk ( p )
[
k =1

Calculate the weight corrections:
∆wij ( p ) =

×xi ( p ) × j ( p )

Update the weights at the hidden neurons:
wij ( p + 1) = wij ( p ) + ∆wij ( p )

© Negnevitsky, Pearson Education, 2005

36
Step 4: Iteration
Increase iteration p by one, go back to Step 2 and
repeat the process until the selected error criterion
is satisfied.
As an example, we may consider the three-layer
back-propagation network. Suppose that the
network is required to perform logical operation
Exclusive-OR. Recall that a single-layer perceptron
could not do this operation. Now we will apply the
three-layer net.

© Negnevitsky, Pearson Education, 2005

37
Three-layer network for solving the
Exclusive-OR operation
−1
θ3
x1

1

w13
w23

3

−1
w35

θ5
5

x2

2

w24
w24

Input
layer

© Negnevitsky, Pearson Education, 2005

y5

w45

4
θ4

−1
Hidden layer

Output
layer

38
 The effect of the threshold applied to a neuron in the
hidden or output layer is represented by its weight, θ,
connected to a fixed input equal to −1.
 The initial weights and threshold levels are set
randomly as follows:
w13 = 0.5, w14 = 0.9, w23 = 0.4, w24 = 1.0, w35 = −1.2,
w45 = 1.1, θ3 = 0.8, θ4 = −0.1 and θ5 = 0.3.

© Negnevitsky, Pearson Education, 2005

39
 We consider a training set where inputs x1 and x2 are
equal to 1 and desired output yd,5 is 0. The actual
outputs of neurons 3 and 4 in the hidden layer are
calculated as
0
0
0
y3 = sigmoid ( x1w13 + x2 w23 − θ3 ) = 1 / 1 + e −(1× .5+1× .4 −1× .8) = 0.5250
0
1
0
y4 = sigmoid ( x1w14 + x2 w24 − θ4 ) = 1 / 1 + e − (1× .9 +1× .0+1× .1) = 0.8808

 Now the actual output of neuron 5 in the output layer
is determined as:
1
1
0
y5 = sigmoid ( y3w35 + y4w45 − θ5 ) = 1/ 1+ e−(−0.5250×.2+0.8808×.1−1× .3) = 0.5097

 Thus, the following error is obtained:
e = yd ,5 − y5 = 0 − 0.5097 = −0.5097
© Negnevitsky, Pearson Education, 2005

40
 The next step is weight training. To update the
weights and threshold levels in our network, we
propagate the error, e, from the output layer
backward to the input layer.
 First, we calculate the error gradient for neuron 5 in
the output layer:
5

= y5 (1 − y5 ) e = 0.5097 × − 0.5097) ×( −0.5097) = −0.1274
(1

 Then we determine the weight corrections assuming
that the learning rate parameter, α, is equal to 0.1:
∆w35 = ×y3 × 5 = 0.1×0.5250 ×(−0.1274) = −0.0067
∆w45 = ×y 4 × 5 = 0.1 ×0.8808 ×(−0.1274 ) = −0.0112
∆θ5 = ×( −1) × 5 = 0.1 ×(−1) ×(−0.1274) = −0.0127
© Negnevitsky, Pearson Education, 2005

41
 Next we calculate the error gradients for neurons 3
and 4 in the hidden layer:
(
(
3 = y3 (1 − y3 ) × 5 ×w35 = 0.5250 ×(1 − 0.5250) × − 0.1274) × − 1.2) = 0.0381
4

= y4 (1 − y4 ) × 5 ×w45 = 0.8808 ×(1 − 0.8808) ×( − 0.127 4) × .1 = −0.0147
1

 We then determine the weight corrections:
∆w13 =
∆w23 =
∆ θ3 =
∆w14 =
∆w24 =
∆θ 4 =

×x1 × 3 = 0.1 × ×0.0381 = 0.0038
1
×x2 × 3 = 0.1 × ×0.0381 = 0.0038
1
×( −1) × 3 = 0.1 ×( −1) ×0.0381 = −0.0038
×x1 × 4 = 0.1 × ×(− 0.0147 ) = −0.0015
1
×x2 × 4 = 0.1 × ×(−0.0147 ) = −0.0015
1
×( −1) × 4 = 0.1 ×( −1) ×( −0 .0147 ) = 0.0015

© Negnevitsky, Pearson Education, 2005

42
 At last, we update all weights and threshold:
w13 = w13 + ∆ w13 = 0 . 5 + 0 . 0038 = 0 .5038
w14 = w14 + ∆ w14 = 0 . 9 − 0 . 0015 = 0 .8985
w 23 = w 23 + ∆ w 23 = 0 . 4 + 0 . 0038 = 0 .4038
w 24 = w 24 + ∆ w 24 = 1 . 0 − 0 . 0015 = 0 .9985
w 35 = w35 + ∆ w35 = − 1 . 2 − 0 . 0067 = − 1 . 2067
w 45 = w 45 + ∆ w 45 = 1 . 1 − 0 . 0112 = 1 .0888
θ 3 = θ 3 + ∆ θ 3 = 0 . 8 − 0 .0038 = 0 . 7962
θ 4 = θ 4 + ∆ θ 4 = − 0 . 1 + 0 . 0015 = − 0 .0985
θ 5 = θ 5 + ∆ θ 5 = 0 . 3 + 0 . 0127 = 0 . 3127

 The training process is repeated until the sum of
squared errors is less than 0.001.
© Negnevitsky, Pearson Education, 2005

43
Learning curve for operation Exclusive-OR
10

Sum-Squared Network Error for 224 Epochs

1

Sum-Squared Error

10 0

10 -1

10 -2

10 -3

10 -4

0

50

© Negnevitsky, Pearson Education, 2005

100
Epoch

150

200
44
Final results of three-layer network learning
Inputs

x1

x2

1
0
1
0

1
1
0
0

Desired
output

Actual
output

yd

y5
0.0155
0.9849
0.9849
0.0175

0
1
1
0

© Negnevitsky, Pearson Education, 2005

e

Sum of
squared
errors

0.0010

45
Network represented by McCulloch-Pitts model
for solving the Exclusive-OR operation
−1
+1.5

x1

1

+1.0

3

−1
−2.0

+1.0

+0.5

5
x2

2

+1.0
+1.0

y5

+1.0

4
+0.5
−1

© Negnevitsky, Pearson Education, 2005

46
Decision boundaries
x2

x2

x2

x1 + x 2 – 1.5 = 0

x 1 + x2 – 0.5 = 0

1

1

1
x1

x1
0

1
(a)

0

1
(b)

x1
0

1
(c)

(a) Decision boundary constructed by hidden neuron 3;
(b) Decision boundary constructed by hidden neuron 4;
(c) Decision boundaries constructed by the complete
three-layer network
© Negnevitsky, Pearson Education, 2005

47
Accelerated learning in multilayer
neural networks
 A multilayer network learns much faster when the
sigmoidal activation function is represented by a
hyperbolic tangent:
2a
tan h
Y
=
−a
1 + e −bX
where a and b are constants.
Suitable values for a and b are:
a = 1.716 and b = 0.667
© Negnevitsky, Pearson Education, 2005

48
 We also can accelerate training by including a
momentum term in the delta rule:

∆w jk ( p) =

×∆w jk ( p − 1) + ×y j ( p ) × k ( p )

where β is a positive number (0 ≤ β < 1) called the
momentum constant. Typically, the momentum
constant is set to 0.95.
This equation is called the generalised delta rule.
© Negnevitsky, Pearson Education, 2005

49
Learning with momentum for operation Exclusive-OR
10

Training for 126 Epochs

2

10 1
10 0
10 -1
10 -2
10 -3
10 -4

0

20

40

60
Epoch

80

100

120

1.5
Learning Rate

1
0.5
0
-0.5
-1

0

20

40

60

80

100

120

140

Epoch
© Negnevitsky, Pearson Education, 2005

50
Learning with adaptive learning rate
To accelerate the convergence and yet avoid the
danger of instability, we can apply two heuristics:

Heuristic 1
If the change of the sum of squared errors has the
same algebraic sign for several consequent epochs,
then the learning rate parameter, α, should be
increased.

Heuristic 2
If the algebraic sign of the change of the sum of
squared errors alternates for several consequent
epochs, then the learning rate parameter, α, should be
decreased.
© Negnevitsky, Pearson Education, 2005

51
 Adapting the learning rate requires some changes
in the back-propagation algorithm.
 If the sum of squared errors at the current epoch
exceeds the previous value by more than a
predefined ratio (typically 1.04), the learning rate
parameter is decreased (typically by multiplying
by 0.7) and new weights and thresholds are
calculated.
 If the error is less than the previous one, the
learning rate is increased (typically by multiplying
by 1.05).
© Negnevitsky, Pearson Education, 2005

52
Learning with adaptive learning rate
Sum-Squared Erro

10

Training for 103 Epochs

2

10 1
10 0
10 -1
10 -2
10 -3
10 -4

0

10

20

30

40

50
60
Epoch

70

80

90

100

1
Learning Rate

0. 8
0. 6
0. 4
0. 2
0

0

20

© Negnevitsky, Pearson Education, 2005

40

60
Epoch

80

100

120

53
Learning with momentum and adaptive learning rate
Sum-Squared Erro

10

Training for 85 Epochs

2

10 1
10 0
10 -1
10 -2
10 -3
10 -4

0

10

0

10

20

30

40
Epoch

50

60

70

80

Learning Rate

2.5
2
1.5
1
0.5
0

20

30

40

50

60

70

80

90

Epoch
© Negnevitsky, Pearson Education, 2005

54
The Hopfield Network
 Neural networks were designed on analogy with
the brain. The brain’s memory, however, works
by association. For example, we can recognise a
familiar face even in an unfamiliar environment
within 100-200 ms. We can also recall a
complete sensory experience, including sounds
and scenes, when we hear only a few bars of
music. The brain routinely associates one thing
with another.

© Negnevitsky, Pearson Education, 2005

55
 Multilayer neural networks trained with the backpropagation algorithm are used for pattern
recognition problems. However, to emulate the
human memory’s associative characteristics we
need a different type of network: a recurrent
neural network.
 A recurrent neural network has feedback loops
from its outputs to its inputs. The presence of
such loops has a profound impact on the learning
capability of the network.

© Negnevitsky, Pearson Education, 2005

56
 The stability of recurrent networks intrigued
several researchers in the 1960s and 1970s.
However, none was able to predict which network
would be stable, and some researchers were
pessimistic about finding a solution at all. The
problem was solved only in 1982, when John
Hopfield formulated the physical principle of
storing information in a dynamically stable
network.

© Negnevitsky, Pearson Education, 2005

57
x1

1

y1

x2

2

y2

xi

i

yi

xn

n

yn

© Negnevitsky, Pearson Education, 2005

Output Signals

Input Signals

Single-layer n-neuron Hopfield network

58
 The Hopfield network uses McCulloch and Pitts
neurons with the sign activation function as its
computing element:
+1, if X > 0
sign 
Y
= −1, if X < 0
 Y , if X = 0


© Negnevitsky, Pearson Education, 2005

59
 The current state of the Hopfield network is
determined by the current outputs of all neurons, y1,
y2, . . ., yn.
Thus, for a single-layer n-neuron network, the state
can be defined by the state vector as:
 y1 
 y 
 2 
Y =




 yn 



© Negnevitsky, Pearson Education, 2005

60
 In the Hopfield network, synaptic weights between
neurons are usually represented in matrix form as
follows:
W=

M

T
YmYm − M I
∑

m=1

where M is the number of states to be memorised
by the network, Ym is the n-dimensional binary
vector, I is n × n identity matrix, and superscript T
denotes matrix transposition.

© Negnevitsky, Pearson Education, 2005

61
Possible states for the three-neuron
Hopfield network
y2
(−1,1, −1)

(1, 1, −1)

(1, 1, 1)

(−1, 1, 1)

y1
0
(1,−1,−1)

(−1,−1,−1)

y3

(−1,−1, 1)

© Negnevitsky, Pearson Education, 2005

(1,−1, 1)
62
 The stable state-vertex is determined by the weight
matrix W, the current input vector X, and the
threshold matrix θ . If the input vector is partially
incorrect or incomplete, the initial state will converge
into the stable state-vertex after a few iterations.
 Suppose, for instance, that our network is required to
memorise two opposite states, (1, 1, 1) and (−1, −1, −1).
Thus,
1
Y1 = 1

1


− 1
Y2 = − 1
 
− 1
 

T
or Y1 = 1 1 1

T
Y2 = − 1 − 1 − 1

where Y1 and Y2 are the three-dimensional vectors.
© Negnevitsky, Pearson Education, 2005

63
 The 3 × 3 identity matrix I is
1 0 0
I = 0 1 0


0 0 1



 Thus, we can now determine the weight matrix as
follows:
1
−1
1 0 0 0
W = 1 1 1 1 + −1 −1 −1 −1 − 2 0 1 0 = 2

 

 
1
−1
0 0 1 2

 

 

2
0
2

2
2

0


 Next, the network is tested by the sequence of input
vectors, X1 and X2, which are equal to the output (or
target) vectors Y1 and Y2, respectively.
© Negnevitsky, Pearson Education, 2005

64
 First, we activate the Hopfield network by applying
the input vector X. Then, we calculate the actual
output vector Y, and finally, we compare the result
with the initial input vector X.
0

Y1 = sign 2
2


2
0
2

2 1 0  1
 1 − 0  = 1
2       
0 1 0  1
     

0

Y2 = sign 2
2


2
0
2

2 −1 0 −1
 −1 − 0 = −1
2      
0 −1 0 −1
      

© Negnevitsky, Pearson Education, 2005

65
 The remaining six states are all unstable. However,
stable states (also called fundamental memories) are
capable of attracting states that are close to them.
 The fundamental memory (1, 1, 1) attracts unstable
states (−1, 1, 1), (1, −1, 1) and (1, 1, −1). Each of
these unstable states represents a single error,
compared to the fundamental memory (1, 1, 1).
 The fundamental memory (−1, −1, −1) attracts
unstable states (−1, −1, 1), (−1, 1, −1) and (1, −1, −1).
 Thus, the Hopfield network can act as an error
correction network.
© Negnevitsky, Pearson Education, 2005

66
Storage capacity of the Hopfield network
 Storage capacity is or the largest number of
fundamental memories that can be stored and
retrieved correctly.
 The maximum number of fundamental memories
Mmax that can be stored in the n-neuron recurrent
network is limited by
M max = 0.15n

© Negnevitsky, Pearson Education, 2005

67
Bidirectional associative memory (BAM)
 The Hopfield network represents an autoassociative
type of memory − it can retrieve a corrupted or
incomplete memory but cannot associate this memory
with another different memory.
 Human memory is essentially associative. One thing
may remind us of another, and that of another, and so
on. We use a chain of mental associations to recover
a lost memory. If we forget where we left an
umbrella, we try to recall where we last had it, what
we were doing, and who we were talking to. We
attempt to establish a chain of associations, and
thereby to restore a lost memory.
© Negnevitsky, Pearson Education, 2005

68
 To associate one memory with another, we need a
recurrent neural network capable of accepting an
input pattern on one set of neurons and producing
a related, but different, output pattern on another
set of neurons.
 Bidirectional associative memory (BAM), first
proposed by Bart Kosko, is a heteroassociative
network. It associates patterns from one set, set A,
to patterns from another set, set B, and vice versa.
Like a Hopfield network, the BAM can generalise
and also produce correct outputs despite corrupted
or incomplete inputs.
© Negnevitsky, Pearson Education, 2005

69
BAM operation
x1(p)

x1(p+1)

1
1

x2 (p)

2

xi (p)

i

y1(p)

2

y2(p)

j

yj(p)

m

xn(p)

1
2

xi(p+1)

i

xn(p+1)
Output
layer

(a) Forward direction.
© Negnevitsky, Pearson Education, 2005

y1(p)

2

y2(p)

j

yj(p)

m

x2(p+1)

ym(p)

n

Input
layer

1

ym(p)

n

Input
layer

Output
layer

(b) Backward direction.
70
The basic idea behind the BAM is to store
pattern pairs so that when n-dimensional vector
X from set A is presented as input, the BAM
recalls m-dimensional vector Y from set B, but
when Y is presented as input, the BAM recalls X.

© Negnevitsky, Pearson Education, 2005

71
 To develop the BAM, we need to create a
correlation matrix for each pattern pair we want to
store. The correlation matrix is the matrix product
of the input vector X, and the transpose of the
output vector YT. The BAM weight matrix is the
sum of all correlation matrices, that is,

W=

M

T
Xm Ym
∑

m=1

where M is the number of pattern pairs to be stored
in the BAM.
© Negnevitsky, Pearson Education, 2005

72
Stability and storage capacity of the BAM
 The BAM is unconditionally stable. This means that
any set of associations can be learned without risk of
instability.

 The maximum number of associations to be stored in
the BAM should not exceed the number of
neurons in the smaller layer.
 The more serious problem with the BAM is
incorrect convergence. The BAM may not
always produce the closest association. In fact, a
stable association may be only slightly related to
the initial input vector.
© Negnevitsky, Pearson Education, 2005

73

Mais conteúdo relacionado

Mais procurados

Artificial Neural Network
Artificial Neural NetworkArtificial Neural Network
Artificial Neural NetworkAtul Krishna
 
Artificial neural network
Artificial neural networkArtificial neural network
Artificial neural networkmustafa aadel
 
Artificial Neural Network
Artificial Neural NetworkArtificial Neural Network
Artificial Neural NetworkPrakash K
 
Activation functions
Activation functionsActivation functions
Activation functionsPRATEEK SAHU
 
Classification using back propagation algorithm
Classification using back propagation algorithmClassification using back propagation algorithm
Classification using back propagation algorithmKIRAN R
 
Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018
Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018
Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018Universitat Politècnica de Catalunya
 
Feedforward neural network
Feedforward neural networkFeedforward neural network
Feedforward neural networkSopheaktra YONG
 
Autoencoders in Deep Learning
Autoencoders in Deep LearningAutoencoders in Deep Learning
Autoencoders in Deep Learningmilad abbasi
 
Spiking neural network: an introduction I
Spiking neural network: an introduction ISpiking neural network: an introduction I
Spiking neural network: an introduction IDalin Zhang
 
Activation function
Activation functionActivation function
Activation functionAstha Jain
 
Principles of soft computing-Associative memory networks
Principles of soft computing-Associative memory networksPrinciples of soft computing-Associative memory networks
Principles of soft computing-Associative memory networksSivagowry Shathesh
 
Machine Learning: Introduction to Neural Networks
Machine Learning: Introduction to Neural NetworksMachine Learning: Introduction to Neural Networks
Machine Learning: Introduction to Neural NetworksFrancesco Collova'
 
Neural network & its applications
Neural network & its applications Neural network & its applications
Neural network & its applications Ahmed_hashmi
 
2.5 backpropagation
2.5 backpropagation2.5 backpropagation
2.5 backpropagationKrish_ver2
 

Mais procurados (20)

Artificial Neural Network
Artificial Neural NetworkArtificial Neural Network
Artificial Neural Network
 
Neural networks introduction
Neural networks introductionNeural networks introduction
Neural networks introduction
 
Artificial neural network
Artificial neural networkArtificial neural network
Artificial neural network
 
Artificial Neural Network
Artificial Neural NetworkArtificial Neural Network
Artificial Neural Network
 
Activation functions
Activation functionsActivation functions
Activation functions
 
Classification using back propagation algorithm
Classification using back propagation algorithmClassification using back propagation algorithm
Classification using back propagation algorithm
 
Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018
Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018
Loss Functions for Deep Learning - Javier Ruiz Hidalgo - UPC Barcelona 2018
 
Artificial neural network
Artificial neural networkArtificial neural network
Artificial neural network
 
Feedforward neural network
Feedforward neural networkFeedforward neural network
Feedforward neural network
 
Autoencoders in Deep Learning
Autoencoders in Deep LearningAutoencoders in Deep Learning
Autoencoders in Deep Learning
 
Spiking neural network: an introduction I
Spiking neural network: an introduction ISpiking neural network: an introduction I
Spiking neural network: an introduction I
 
Activation function
Activation functionActivation function
Activation function
 
Principles of soft computing-Associative memory networks
Principles of soft computing-Associative memory networksPrinciples of soft computing-Associative memory networks
Principles of soft computing-Associative memory networks
 
Neural Networks
Neural NetworksNeural Networks
Neural Networks
 
Machine Learning: Introduction to Neural Networks
Machine Learning: Introduction to Neural NetworksMachine Learning: Introduction to Neural Networks
Machine Learning: Introduction to Neural Networks
 
Neural network & its applications
Neural network & its applications Neural network & its applications
Neural network & its applications
 
Adaptive Resonance Theory (ART)
Adaptive Resonance Theory (ART)Adaptive Resonance Theory (ART)
Adaptive Resonance Theory (ART)
 
Mc culloch pitts neuron
Mc culloch pitts neuronMc culloch pitts neuron
Mc culloch pitts neuron
 
2.5 backpropagation
2.5 backpropagation2.5 backpropagation
2.5 backpropagation
 
HOPFIELD NETWORK
HOPFIELD NETWORKHOPFIELD NETWORK
HOPFIELD NETWORK
 

Destaque

Du binary signalling
Du binary signallingDu binary signalling
Du binary signallingsrkrishna341
 
Correlative level coding
Correlative level codingCorrelative level coding
Correlative level codingsrkrishna341
 
Introduction to Neural networks (under graduate course) Lecture 1 of 9
Introduction to Neural networks (under graduate course) Lecture 1 of 9Introduction to Neural networks (under graduate course) Lecture 1 of 9
Introduction to Neural networks (under graduate course) Lecture 1 of 9Randa Elanwar
 
Neural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) AlgorithmNeural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) AlgorithmMostafa G. M. Mostafa
 
NEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMS
NEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMSNEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMS
NEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMSESCOM
 
Recurrent Neural Networks. Part 1: Theory
Recurrent Neural Networks. Part 1: TheoryRecurrent Neural Networks. Part 1: Theory
Recurrent Neural Networks. Part 1: TheoryAndrii Gakhov
 
Artificial Neural Networks Lect3: Neural Network Learning rules
Artificial Neural Networks Lect3: Neural Network Learning rulesArtificial Neural Networks Lect3: Neural Network Learning rules
Artificial Neural Networks Lect3: Neural Network Learning rulesMohammed Bennamoun
 
Artificial Neural Network
Artificial Neural Network Artificial Neural Network
Artificial Neural Network Iman Ardekani
 
Neural network
Neural networkNeural network
Neural networkSilicon
 

Destaque (10)

Du binary signalling
Du binary signallingDu binary signalling
Du binary signalling
 
Correlative level coding
Correlative level codingCorrelative level coding
Correlative level coding
 
Introduction to Neural networks (under graduate course) Lecture 1 of 9
Introduction to Neural networks (under graduate course) Lecture 1 of 9Introduction to Neural networks (under graduate course) Lecture 1 of 9
Introduction to Neural networks (under graduate course) Lecture 1 of 9
 
Neural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) AlgorithmNeural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) Algorithm
 
NEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMS
NEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMSNEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMS
NEURAL NETWORK Widrow-Hoff Learning Adaline Hagan LMS
 
Recurrent Neural Networks. Part 1: Theory
Recurrent Neural Networks. Part 1: TheoryRecurrent Neural Networks. Part 1: Theory
Recurrent Neural Networks. Part 1: Theory
 
Artificial Neural Networks Lect3: Neural Network Learning rules
Artificial Neural Networks Lect3: Neural Network Learning rulesArtificial Neural Networks Lect3: Neural Network Learning rules
Artificial Neural Networks Lect3: Neural Network Learning rules
 
Hopfield Networks
Hopfield NetworksHopfield Networks
Hopfield Networks
 
Artificial Neural Network
Artificial Neural Network Artificial Neural Network
Artificial Neural Network
 
Neural network
Neural networkNeural network
Neural network
 

Semelhante a neural networks

2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…Dongseo University
 
SOFT COMPUTERING TECHNICS -Unit 1
SOFT COMPUTERING TECHNICS -Unit 1SOFT COMPUTERING TECHNICS -Unit 1
SOFT COMPUTERING TECHNICS -Unit 1sravanthi computers
 
lecture07.ppt
lecture07.pptlecture07.ppt
lecture07.pptbutest
 
ACUMENS ON NEURAL NET AKG 20 7 23.pptx
ACUMENS ON NEURAL NET AKG 20 7 23.pptxACUMENS ON NEURAL NET AKG 20 7 23.pptx
ACUMENS ON NEURAL NET AKG 20 7 23.pptxgnans Kgnanshek
 
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…Dongseo University
 
Soft Computing-173101
Soft Computing-173101Soft Computing-173101
Soft Computing-173101AMIT KUMAR
 
Artificial neural networks (2)
Artificial neural networks (2)Artificial neural networks (2)
Artificial neural networks (2)sai anjaneya
 
Csss2010 20100803-kanevski-lecture2
Csss2010 20100803-kanevski-lecture2Csss2010 20100803-kanevski-lecture2
Csss2010 20100803-kanevski-lecture2hasan_elektro
 
Neural network
Neural networkNeural network
Neural networkmarada0033
 
JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience
JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience
JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience hirokazutanaka
 
Neural Networks Ver1
Neural  Networks  Ver1Neural  Networks  Ver1
Neural Networks Ver1ncct
 
Artificial Neural Network
Artificial Neural NetworkArtificial Neural Network
Artificial Neural NetworkRenas Rekany
 
Multilayer Backpropagation Neural Networks for Implementation of Logic Gates
Multilayer Backpropagation Neural Networks for Implementation of Logic GatesMultilayer Backpropagation Neural Networks for Implementation of Logic Gates
Multilayer Backpropagation Neural Networks for Implementation of Logic GatesIJCSES Journal
 

Semelhante a neural networks (20)

2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
 
SOFT COMPUTERING TECHNICS -Unit 1
SOFT COMPUTERING TECHNICS -Unit 1SOFT COMPUTERING TECHNICS -Unit 1
SOFT COMPUTERING TECHNICS -Unit 1
 
lecture07.ppt
lecture07.pptlecture07.ppt
lecture07.ppt
 
10-Perceptron.pdf
10-Perceptron.pdf10-Perceptron.pdf
10-Perceptron.pdf
 
ACUMENS ON NEURAL NET AKG 20 7 23.pptx
ACUMENS ON NEURAL NET AKG 20 7 23.pptxACUMENS ON NEURAL NET AKG 20 7 23.pptx
ACUMENS ON NEURAL NET AKG 20 7 23.pptx
 
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
2013-1 Machine Learning Lecture 04 - Michael Negnevitsky - Artificial neur…
 
Lec 5
Lec 5Lec 5
Lec 5
 
Soft Computing-173101
Soft Computing-173101Soft Computing-173101
Soft Computing-173101
 
Perceptron
PerceptronPerceptron
Perceptron
 
Artificial neural networks (2)
Artificial neural networks (2)Artificial neural networks (2)
Artificial neural networks (2)
 
Csss2010 20100803-kanevski-lecture2
Csss2010 20100803-kanevski-lecture2Csss2010 20100803-kanevski-lecture2
Csss2010 20100803-kanevski-lecture2
 
Lec10
Lec10Lec10
Lec10
 
Neural network
Neural networkNeural network
Neural network
 
JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience
JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience
JAISTサマースクール2016「脳を知るための理論」講義04 Neural Networks and Neuroscience
 
Neural Networks Ver1
Neural  Networks  Ver1Neural  Networks  Ver1
Neural Networks Ver1
 
Unit 2
Unit 2Unit 2
Unit 2
 
Artificial Neural Network
Artificial Neural NetworkArtificial Neural Network
Artificial Neural Network
 
Multilayer Backpropagation Neural Networks for Implementation of Logic Gates
Multilayer Backpropagation Neural Networks for Implementation of Logic GatesMultilayer Backpropagation Neural Networks for Implementation of Logic Gates
Multilayer Backpropagation Neural Networks for Implementation of Logic Gates
 
Nn3
Nn3Nn3
Nn3
 
19_Learning.ppt
19_Learning.ppt19_Learning.ppt
19_Learning.ppt
 

Mais de Institute of Technology Telkom

Mais de Institute of Technology Telkom (20)

Econopysics
EconopysicsEconopysics
Econopysics
 
Science and religion 100622120615-phpapp01
Science and religion 100622120615-phpapp01Science and religion 100622120615-phpapp01
Science and religion 100622120615-phpapp01
 
Konvergensi sains dan_spiritualitas
Konvergensi sains dan_spiritualitasKonvergensi sains dan_spiritualitas
Konvergensi sains dan_spiritualitas
 
Matematika arah kiblat mikrajuddin abdullah 2017
Matematika arah kiblat   mikrajuddin abdullah 2017Matematika arah kiblat   mikrajuddin abdullah 2017
Matematika arah kiblat mikrajuddin abdullah 2017
 
Iau solar effects 2005
Iau solar effects 2005Iau solar effects 2005
Iau solar effects 2005
 
Hfmsilri2jun14
Hfmsilri2jun14Hfmsilri2jun14
Hfmsilri2jun14
 
Fisika komputasi
Fisika komputasiFisika komputasi
Fisika komputasi
 
Computer Aided Process Planning
Computer Aided Process PlanningComputer Aided Process Planning
Computer Aided Process Planning
 
Archimedes
ArchimedesArchimedes
Archimedes
 
Web and text
Web and textWeb and text
Web and text
 
Web data mining
Web data miningWeb data mining
Web data mining
 
Time series Forecasting using svm
Time series Forecasting using  svmTime series Forecasting using  svm
Time series Forecasting using svm
 
Timeseries forecasting
Timeseries forecastingTimeseries forecasting
Timeseries forecasting
 
Fuzzy logic
Fuzzy logicFuzzy logic
Fuzzy logic
 
World population 1950--2050
World population 1950--2050World population 1950--2050
World population 1950--2050
 
Artificial neural networks
Artificial neural networks Artificial neural networks
Artificial neural networks
 
002 ray modeling dynamic systems
002 ray modeling dynamic systems002 ray modeling dynamic systems
002 ray modeling dynamic systems
 
002 ray modeling dynamic systems
002 ray modeling dynamic systems002 ray modeling dynamic systems
002 ray modeling dynamic systems
 
System dynamics majors fair
System dynamics majors fairSystem dynamics majors fair
System dynamics majors fair
 
System dynamics math representation
System dynamics math representationSystem dynamics math representation
System dynamics math representation
 

Último

Prescribed medication order and communication skills.pptx
Prescribed medication order and communication skills.pptxPrescribed medication order and communication skills.pptx
Prescribed medication order and communication skills.pptxraviapr7
 
How to Use api.constrains ( ) in Odoo 17
How to Use api.constrains ( ) in Odoo 17How to Use api.constrains ( ) in Odoo 17
How to Use api.constrains ( ) in Odoo 17Celine George
 
CapTechU Doctoral Presentation -March 2024 slides.pptx
CapTechU Doctoral Presentation -March 2024 slides.pptxCapTechU Doctoral Presentation -March 2024 slides.pptx
CapTechU Doctoral Presentation -March 2024 slides.pptxCapitolTechU
 
Presentation on the Basics of Writing. Writing a Paragraph
Presentation on the Basics of Writing. Writing a ParagraphPresentation on the Basics of Writing. Writing a Paragraph
Presentation on the Basics of Writing. Writing a ParagraphNetziValdelomar1
 
How to Add Existing Field in One2Many Tree View in Odoo 17
How to Add Existing Field in One2Many Tree View in Odoo 17How to Add Existing Field in One2Many Tree View in Odoo 17
How to Add Existing Field in One2Many Tree View in Odoo 17Celine George
 
How to Add a New Field in Existing Kanban View in Odoo 17
How to Add a New Field in Existing Kanban View in Odoo 17How to Add a New Field in Existing Kanban View in Odoo 17
How to Add a New Field in Existing Kanban View in Odoo 17Celine George
 
Practical Research 1 Lesson 9 Scope and delimitation.pptx
Practical Research 1 Lesson 9 Scope and delimitation.pptxPractical Research 1 Lesson 9 Scope and delimitation.pptx
Practical Research 1 Lesson 9 Scope and delimitation.pptxKatherine Villaluna
 
The basics of sentences session 10pptx.pptx
The basics of sentences session 10pptx.pptxThe basics of sentences session 10pptx.pptx
The basics of sentences session 10pptx.pptxheathfieldcps1
 
2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx
2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx
2024.03.23 What do successful readers do - Sandy Millin for PARK.pptxSandy Millin
 
General views of Histopathology and step
General views of Histopathology and stepGeneral views of Histopathology and step
General views of Histopathology and stepobaje godwin sunday
 
Diploma in Nursing Admission Test Question Solution 2023.pdf
Diploma in Nursing Admission Test Question Solution 2023.pdfDiploma in Nursing Admission Test Question Solution 2023.pdf
Diploma in Nursing Admission Test Question Solution 2023.pdfMohonDas
 
3.21.24 The Origins of Black Power.pptx
3.21.24  The Origins of Black Power.pptx3.21.24  The Origins of Black Power.pptx
3.21.24 The Origins of Black Power.pptxmary850239
 
How to Add a many2many Relational Field in Odoo 17
How to Add a many2many Relational Field in Odoo 17How to Add a many2many Relational Field in Odoo 17
How to Add a many2many Relational Field in Odoo 17Celine George
 
Ultra structure and life cycle of Plasmodium.pptx
Ultra structure and life cycle of Plasmodium.pptxUltra structure and life cycle of Plasmodium.pptx
Ultra structure and life cycle of Plasmodium.pptxDr. Asif Anas
 
Clinical Pharmacy Introduction to Clinical Pharmacy, Concept of clinical pptx
Clinical Pharmacy  Introduction to Clinical Pharmacy, Concept of clinical pptxClinical Pharmacy  Introduction to Clinical Pharmacy, Concept of clinical pptx
Clinical Pharmacy Introduction to Clinical Pharmacy, Concept of clinical pptxraviapr7
 
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...Nguyen Thanh Tu Collection
 
The Singapore Teaching Practice document
The Singapore Teaching Practice documentThe Singapore Teaching Practice document
The Singapore Teaching Practice documentXsasf Sfdfasd
 
How to Make a Field read-only in Odoo 17
How to Make a Field read-only in Odoo 17How to Make a Field read-only in Odoo 17
How to Make a Field read-only in Odoo 17Celine George
 
Easter in the USA presentation by Chloe.
Easter in the USA presentation by Chloe.Easter in the USA presentation by Chloe.
Easter in the USA presentation by Chloe.EnglishCEIPdeSigeiro
 

Último (20)

Prescribed medication order and communication skills.pptx
Prescribed medication order and communication skills.pptxPrescribed medication order and communication skills.pptx
Prescribed medication order and communication skills.pptx
 
How to Use api.constrains ( ) in Odoo 17
How to Use api.constrains ( ) in Odoo 17How to Use api.constrains ( ) in Odoo 17
How to Use api.constrains ( ) in Odoo 17
 
CapTechU Doctoral Presentation -March 2024 slides.pptx
CapTechU Doctoral Presentation -March 2024 slides.pptxCapTechU Doctoral Presentation -March 2024 slides.pptx
CapTechU Doctoral Presentation -March 2024 slides.pptx
 
Presentation on the Basics of Writing. Writing a Paragraph
Presentation on the Basics of Writing. Writing a ParagraphPresentation on the Basics of Writing. Writing a Paragraph
Presentation on the Basics of Writing. Writing a Paragraph
 
How to Add Existing Field in One2Many Tree View in Odoo 17
How to Add Existing Field in One2Many Tree View in Odoo 17How to Add Existing Field in One2Many Tree View in Odoo 17
How to Add Existing Field in One2Many Tree View in Odoo 17
 
How to Add a New Field in Existing Kanban View in Odoo 17
How to Add a New Field in Existing Kanban View in Odoo 17How to Add a New Field in Existing Kanban View in Odoo 17
How to Add a New Field in Existing Kanban View in Odoo 17
 
Practical Research 1 Lesson 9 Scope and delimitation.pptx
Practical Research 1 Lesson 9 Scope and delimitation.pptxPractical Research 1 Lesson 9 Scope and delimitation.pptx
Practical Research 1 Lesson 9 Scope and delimitation.pptx
 
Personal Resilience in Project Management 2 - TV Edit 1a.pdf
Personal Resilience in Project Management 2 - TV Edit 1a.pdfPersonal Resilience in Project Management 2 - TV Edit 1a.pdf
Personal Resilience in Project Management 2 - TV Edit 1a.pdf
 
The basics of sentences session 10pptx.pptx
The basics of sentences session 10pptx.pptxThe basics of sentences session 10pptx.pptx
The basics of sentences session 10pptx.pptx
 
2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx
2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx
2024.03.23 What do successful readers do - Sandy Millin for PARK.pptx
 
General views of Histopathology and step
General views of Histopathology and stepGeneral views of Histopathology and step
General views of Histopathology and step
 
Diploma in Nursing Admission Test Question Solution 2023.pdf
Diploma in Nursing Admission Test Question Solution 2023.pdfDiploma in Nursing Admission Test Question Solution 2023.pdf
Diploma in Nursing Admission Test Question Solution 2023.pdf
 
3.21.24 The Origins of Black Power.pptx
3.21.24  The Origins of Black Power.pptx3.21.24  The Origins of Black Power.pptx
3.21.24 The Origins of Black Power.pptx
 
How to Add a many2many Relational Field in Odoo 17
How to Add a many2many Relational Field in Odoo 17How to Add a many2many Relational Field in Odoo 17
How to Add a many2many Relational Field in Odoo 17
 
Ultra structure and life cycle of Plasmodium.pptx
Ultra structure and life cycle of Plasmodium.pptxUltra structure and life cycle of Plasmodium.pptx
Ultra structure and life cycle of Plasmodium.pptx
 
Clinical Pharmacy Introduction to Clinical Pharmacy, Concept of clinical pptx
Clinical Pharmacy  Introduction to Clinical Pharmacy, Concept of clinical pptxClinical Pharmacy  Introduction to Clinical Pharmacy, Concept of clinical pptx
Clinical Pharmacy Introduction to Clinical Pharmacy, Concept of clinical pptx
 
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...
 
The Singapore Teaching Practice document
The Singapore Teaching Practice documentThe Singapore Teaching Practice document
The Singapore Teaching Practice document
 
How to Make a Field read-only in Odoo 17
How to Make a Field read-only in Odoo 17How to Make a Field read-only in Odoo 17
How to Make a Field read-only in Odoo 17
 
Easter in the USA presentation by Chloe.
Easter in the USA presentation by Chloe.Easter in the USA presentation by Chloe.
Easter in the USA presentation by Chloe.
 

neural networks

  • 1. Lecture 7 Artificial neural networks: Supervised learning  Introduction, or how the brain works  The neuron as a simple computing element  The perceptron  Multilayer neural networks  Accelerated learning in multilayer neural networks  The Hopfield network  Bidirectional associative memories (BAM)  Summary © Negnevitsky, Pearson Education, 2005 1
  • 2. Introduction, or how the brain works Machine learning involves adaptive mechanisms that enable computers to learn from experience, learn by example and learn by analogy. Learning capabilities can improve the performance of an intelligent system over time. The most popular approaches to machine learning are artificial neural networks and genetic algorithms. This lecture is dedicated to neural networks. © Negnevitsky, Pearson Education, 2005 2
  • 3.  A neural network can be defined as a model of reasoning based on the human brain. The brain consists of a densely interconnected set of nerve cells, or basic information-processing units, called neurons.  The human brain incorporates nearly 10 billion neurons and 60 trillion connections, synapses, between them. By using multiple neurons simultaneously, the brain can perform its functions much faster than the fastest computers in existence today. © Negnevitsky, Pearson Education, 2005 3
  • 4.  Each neuron has a very simple structure, but an army of such elements constitutes a tremendous processing power.  A neuron consists of a cell body, soma, a number of fibers called dendrites, and a single long fiber called the axon. © Negnevitsky, Pearson Education, 2005 4
  • 6.  Our brain can be considered as a highly complex, non-linear and parallel information-processing system.  Information is stored and processed in a neural network simultaneously throughout the whole network, rather than at specific locations. In other words, in neural networks, both data and its processing are global rather than local.  Learning is a fundamental and essential characteristic of biological neural networks. The ease with which they can learn led to attempts to emulate a biological neural network in a computer. © Negnevitsky, Pearson Education, 2005 6
  • 7.  An artificial neural network consists of a number of very simple processors, also called neurons, which are analogous to the biological neurons in the brain.  The neurons are connected by weighted links passing signals from one neuron to another.  The output signal is transmitted through the neuron’s outgoing connection. The outgoing connection splits into a number of branches that transmit the same signal. The outgoing branches terminate at the incoming connections of other neurons in the network. © Negnevitsky, Pearson Education, 2005 7
  • 8. Input Signals Output Signals Architecture of a typical artificial neural network Middle Layer Input Layer © Negnevitsky, Pearson Education, 2005 Output Layer 8
  • 9. Analogy between biological and artificial neural networks Biological Neural Network Soma Dendrite Axon Synapse © Negnevitsky, Pearson Education, 2005 Artificial Neural Network Neuron Input Output Weight 9
  • 10. The neuron as a simple computing element Diagram of a neuron Input Signals x1 x2 xn Weights Output Signals Y w1 w2 wn © Negnevitsky, Pearson Education, 2005 Neuron Y Y Y 10
  • 11.  The neuron computes the weighted sum of the input signals and compares the result with a threshold value, θ. If the net input is less than the threshold, the neuron output is –1. But if the net input is greater than or equal to the threshold, the neuron becomes activated and its output attains a value +1.  The neuron uses the following transfer or activation function: n X = ∑ xi wi i =1 +1, if X ≥ θ Y = −1, if X < θ  This type of activation function is called a sign function. © Negnevitsky, Pearson Education, 2005 11
  • 12. Activation functions of a neuron Step function Sign function Sigmoid function Linear function Y Y Y Y +1 +1 1 1 0 X 0 X -1 -1 0 -1 1 , step=  , if X ≥ 0 Y sign = +1 if X ≥ 0 Y sigmoid= Y   0, if X < 0 −1, if X < 0 © Negnevitsky, Pearson Education, 2005 X 0 X -1 1 1 + e− X Y linear= X 12
  • 13. Can a single neuron learn a task?  In 1958, Frank Rosenblatt introduced a training algorithm that provided the first procedure for training a simple ANN: a perceptron.  The perceptron is the simplest form of a neural network. It consists of a single neuron with adjustable synaptic weights and a hard limiter. © Negnevitsky, Pearson Education, 2005 13
  • 15. The Perceptron  The operation of Rosenblatt’s perceptron is based on the McCulloch and Pitts neuron model. The model consists of a linear combiner followed by a hard limiter.  The weighted sum of the inputs is applied to the hard limiter, which produces an output equal to +1 if its input is positive and −1 if it is negative. © Negnevitsky, Pearson Education, 2005 15
  • 16.  The aim of the perceptron is to classify inputs, x1, x2, . . ., xn, into one of two classes, say A1 and A2.  In the case of an elementary perceptron, the ndimensional space is divided by a hyperplane into two decision regions. The hyperplane is defined by the linearly separable function: n ∑ xi wi − θ = 0 i =1 © Negnevitsky, Pearson Education, 2005 16
  • 17. Linear separability in the perceptrons x2 x2 Class A 1 1 1 Class A 2 2 x1 x1 2 x 1w 1 + x 2w 2 −θ = 0 (a) Two-input perceptron. © Negnevitsky, Pearson Education, 2005 x3 x1w 1 + x2w 2 + x3w 3 −θ = 0 (b) Three-input perceptron. 17
  • 18. How does the perceptron learn its classification tasks? This is done by making small adjustments in the weights to reduce the difference between the actual and desired outputs of the perceptron. The initial weights are randomly assigned, usually in the range [−0.5, 0.5], and then updated to obtain the output consistent with the training examples. © Negnevitsky, Pearson Education, 2005 18
  • 19.  If at iteration p, the actual output is Y(p) and the desired output is Yd (p), then the error is given by: e( p) = Yd ( p) − Y ( p) where p = 1, 2, 3, . . . Iteration p here refers to the pth training example presented to the perceptron.  If the error, e(p), is positive, we need to increase perceptron output Y(p), but if it is negative, we need to decrease Y(p). © Negnevitsky, Pearson Education, 2005 19
  • 20. The perceptron learning rule wi ( p + 1) = wi ( p ) +  . xi ( p ) . e( p ) × where p = 1, 2, 3, . . . α is the learning rate, a positive constant less than unity. The perceptron learning rule was first proposed by Rosenblatt in 1960. Using this rule we can derive the perceptron training algorithm for classification tasks. © Negnevitsky, Pearson Education, 2005 20
  • 21. Perceptron’s training algorithm Step 1: Initialisation Set initial weights w1, w2,…, wn and threshold θ to random numbers in the range [−0.5, 0.5]. If the error, e(p), is positive, we need to increase perceptron output Y(p), but if it is negative, we need to decrease Y(p). © Negnevitsky, Pearson Education, 2005 21
  • 22. Perceptron’s training algorithm (continued) Step 2: Activation Activate the perceptron by applying inputs x1(p), x2(p),…, xn(p) and desired output Yd (p). Calculate the actual output at iteration p = 1  n  Y ( p ) = step  ∑ x i ( p ) w i ( p ) − θ   i =1    where n is the number of the perceptron inputs, and step is a step activation function. © Negnevitsky, Pearson Education, 2005 22
  • 23. Perceptron’s training algorithm (continued) Step 3: Weight training Update the weights of the perceptron wi ( p + 1) = wi ( p) + ∆wi ( p) where ∆wi(p) is the weight correction at iteration p. The weight correction is computed by the delta rule: . ∆wi ( p) = α ×xi ( p ) ×e( p) Step 4: Iteration Increase iteration p by one, go back to Step 2 and repeat the process until convergence. © Negnevitsky, Pearson Education, 2005 23
  • 24. Example of perceptron learning: the logical operation AND Epoch Inputs Desired output Yd Initial weights w1 w2 Actual output Y Error Final weights w1 w2 x1 x2 1 0 0 1 1 0 1 0 1 0 0 0 1 0.3 0.3 0.3 0.2 − 0.1 − 0.1 − 0.1 − 0.1 0 0 1 0 0 0 −1 1 0.3 0.3 0.2 0.3 − 0.1 − 0.1 − 0.1 0.0 2 0 0 1 1 0 1 0 1 0 0 0 1 0.3 0.3 0.3 0.2 0.0 0.0 0.0 0.0 0 0 1 1 0 0 −1 0 0.3 0.3 0.2 0.2 0.0 0.0 0.0 0.0 3 0 0 1 1 0 1 0 1 0 0 0 1 0.2 0.2 0.2 0.1 0.0 0.0 0.0 0.0 0 0 1 0 0 0 −1 1 0.2 0.2 0.1 0.2 0.0 0.0 0.0 0.1 4 0 0 1 1 0 1 0 1 0 0 0 1 0.2 0.2 0.2 0.1 0.1 0.1 0.1 0.1 0 0 1 1 0 0 −1 0 0.2 0.2 0.1 0.1 0.1 0.1 0.1 0.1 5 0 0 1 1 0 1 0 1 0 0 0 1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0 0 0 1 0 0 0 0 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 Threshold: θ = 0.2; learning rate: © Negnevitsky, Pearson Education, 2005 e = 0.1 24
  • 25. Two-dimensional plots of basic logical operations x2 x2 x2 1 1 1 x1 x1 0 1 (a) AND (x1 ∩ x2) 0 1 (b) OR (x 1 ∪ x 2 ) x1 0 1 (c) Ex cl us iv e- OR (x 1 ⊕ x 2 ) A perceptron can learn the operations AND and OR, but not Exclusive-OR. © Negnevitsky, Pearson Education, 2005 25
  • 26. Multilayer neural networks  A multilayer perceptron is a feedforward neural network with one or more hidden layers.  The network consists of an input layer of source neurons, at least one middle or hidden layer of computational neurons, and an output layer of computational neurons.  The input signals are propagated in a forward direction on a layer-by-layer basis. © Negnevitsky, Pearson Education, 2005 26
  • 27. Input Signals Output Signals Multilayer perceptron with two hidden layers Input layer First hidden layer © Negnevitsky, Pearson Education, 2005 Second hidden layer Output layer 27
  • 28. What does the middle layer hide?  A hidden layer “hides” its desired output. Neurons in the hidden layer cannot be observed through the input/output behaviour of the network. There is no obvious way to know what the desired output of the hidden layer should be.  Commercial ANNs incorporate three and sometimes four layers, including one or two hidden layers. Each layer can contain from 10 to 1000 neurons. Experimental neural networks may have five or even six layers, including three or four hidden layers, and utilise millions of neurons. © Negnevitsky, Pearson Education, 2005 28
  • 29. Back-propagation neural network  Learning in a multilayer network proceeds the same way as for a perceptron.  A training set of input patterns is presented to the network.  The network computes its output pattern, and if there is an error − or in other words a difference between actual and desired output patterns − the weights are adjusted to reduce this error. © Negnevitsky, Pearson Education, 2005 29
  • 30.  In a back-propagation neural network, the learning algorithm has two phases.  First, a training input pattern is presented to the network input layer. The network propagates the input pattern from layer to layer until the output pattern is generated by the output layer.  If this pattern is different from the desired output, an error is calculated and then propagated backwards through the network from the output layer to the input layer. The weights are modified as the error is propagated. © Negnevitsky, Pearson Education, 2005 30
  • 31. Three-layer back-propagation neural network Input signals x1 x2 xi 1 1 2 y2 k yk l yl 1 2 2 i y1 wij j wjk m xn n Input layer Hidden layer Output layer Error signals © Negnevitsky, Pearson Education, 2005 31
  • 32. The back-propagation training algorithm Step 1: Initialisation Set all the weights and threshold levels of the network to random numbers uniformly distributed inside a small range:  2.4 2.4  −  F , + F ÷ ÷ i i   where Fi is the total number of inputs of neuron i in the network. The weight initialisation is done on a neuron-by-neuron basis. © Negnevitsky, Pearson Education, 2005 32
  • 33. Step 2: Activation Activate the back-propagation neural network by applying inputs x1(p), x2(p),…, xn(p) and desired outputs yd,1(p), yd,2(p),…, yd,n(p). (a) Calculate the actual outputs of the neurons in the hidden layer:  n  y j ( p ) = sigmoid  ∑ xi ( p ) ×wij ( p ) − θ j   i =1    where n is the number of inputs of neuron j in the hidden layer, and sigmoid is the sigmoid activation function. © Negnevitsky, Pearson Education, 2005 33
  • 34. Step 2 : Activation (continued) (b) Calculate the actual outputs of the neurons in the output layer: m  y k ( p ) = sigmoid  ∑ x jk ( p ) ×w jk ( p ) − θ k   j =1    where m is the number of inputs of neuron k in the output layer. © Negnevitsky, Pearson Education, 2005 34
  • 35. Step 3: Weight training Update the weights in the back-propagation network propagating backward the errors associated with output neurons. (a) Calculate the error gradient for the neurons in the output layer: k ( p) = yk ( p) ×1 − y k ( p ) ×ek ( p ) where ek ( p ) = yd ,k ( p ) − yk ( p ) Calculate the weight corrections: ∆w jk ( p) = ×y j ( p) × k ( p) Update the weights at the output neurons: w jk ( p + 1) = w jk ( p ) + ∆w jk ( p ) © Negnevitsky, Pearson Education, 2005 35
  • 36. Step 3: Weight training (continued) (b) Calculate the error gradient for the neurons in the hidden layer: l j ( p) = y j ( p ) × 1 − y j ( p )] ×∑ k ( p ) w jk ( p ) [ k =1 Calculate the weight corrections: ∆wij ( p ) = ×xi ( p ) × j ( p ) Update the weights at the hidden neurons: wij ( p + 1) = wij ( p ) + ∆wij ( p ) © Negnevitsky, Pearson Education, 2005 36
  • 37. Step 4: Iteration Increase iteration p by one, go back to Step 2 and repeat the process until the selected error criterion is satisfied. As an example, we may consider the three-layer back-propagation network. Suppose that the network is required to perform logical operation Exclusive-OR. Recall that a single-layer perceptron could not do this operation. Now we will apply the three-layer net. © Negnevitsky, Pearson Education, 2005 37
  • 38. Three-layer network for solving the Exclusive-OR operation −1 θ3 x1 1 w13 w23 3 −1 w35 θ5 5 x2 2 w24 w24 Input layer © Negnevitsky, Pearson Education, 2005 y5 w45 4 θ4 −1 Hidden layer Output layer 38
  • 39.  The effect of the threshold applied to a neuron in the hidden or output layer is represented by its weight, θ, connected to a fixed input equal to −1.  The initial weights and threshold levels are set randomly as follows: w13 = 0.5, w14 = 0.9, w23 = 0.4, w24 = 1.0, w35 = −1.2, w45 = 1.1, θ3 = 0.8, θ4 = −0.1 and θ5 = 0.3. © Negnevitsky, Pearson Education, 2005 39
  • 40.  We consider a training set where inputs x1 and x2 are equal to 1 and desired output yd,5 is 0. The actual outputs of neurons 3 and 4 in the hidden layer are calculated as 0 0 0 y3 = sigmoid ( x1w13 + x2 w23 − θ3 ) = 1 / 1 + e −(1× .5+1× .4 −1× .8) = 0.5250 0 1 0 y4 = sigmoid ( x1w14 + x2 w24 − θ4 ) = 1 / 1 + e − (1× .9 +1× .0+1× .1) = 0.8808  Now the actual output of neuron 5 in the output layer is determined as: 1 1 0 y5 = sigmoid ( y3w35 + y4w45 − θ5 ) = 1/ 1+ e−(−0.5250×.2+0.8808×.1−1× .3) = 0.5097  Thus, the following error is obtained: e = yd ,5 − y5 = 0 − 0.5097 = −0.5097 © Negnevitsky, Pearson Education, 2005 40
  • 41.  The next step is weight training. To update the weights and threshold levels in our network, we propagate the error, e, from the output layer backward to the input layer.  First, we calculate the error gradient for neuron 5 in the output layer: 5 = y5 (1 − y5 ) e = 0.5097 × − 0.5097) ×( −0.5097) = −0.1274 (1  Then we determine the weight corrections assuming that the learning rate parameter, α, is equal to 0.1: ∆w35 = ×y3 × 5 = 0.1×0.5250 ×(−0.1274) = −0.0067 ∆w45 = ×y 4 × 5 = 0.1 ×0.8808 ×(−0.1274 ) = −0.0112 ∆θ5 = ×( −1) × 5 = 0.1 ×(−1) ×(−0.1274) = −0.0127 © Negnevitsky, Pearson Education, 2005 41
  • 42.  Next we calculate the error gradients for neurons 3 and 4 in the hidden layer: ( ( 3 = y3 (1 − y3 ) × 5 ×w35 = 0.5250 ×(1 − 0.5250) × − 0.1274) × − 1.2) = 0.0381 4 = y4 (1 − y4 ) × 5 ×w45 = 0.8808 ×(1 − 0.8808) ×( − 0.127 4) × .1 = −0.0147 1  We then determine the weight corrections: ∆w13 = ∆w23 = ∆ θ3 = ∆w14 = ∆w24 = ∆θ 4 = ×x1 × 3 = 0.1 × ×0.0381 = 0.0038 1 ×x2 × 3 = 0.1 × ×0.0381 = 0.0038 1 ×( −1) × 3 = 0.1 ×( −1) ×0.0381 = −0.0038 ×x1 × 4 = 0.1 × ×(− 0.0147 ) = −0.0015 1 ×x2 × 4 = 0.1 × ×(−0.0147 ) = −0.0015 1 ×( −1) × 4 = 0.1 ×( −1) ×( −0 .0147 ) = 0.0015 © Negnevitsky, Pearson Education, 2005 42
  • 43.  At last, we update all weights and threshold: w13 = w13 + ∆ w13 = 0 . 5 + 0 . 0038 = 0 .5038 w14 = w14 + ∆ w14 = 0 . 9 − 0 . 0015 = 0 .8985 w 23 = w 23 + ∆ w 23 = 0 . 4 + 0 . 0038 = 0 .4038 w 24 = w 24 + ∆ w 24 = 1 . 0 − 0 . 0015 = 0 .9985 w 35 = w35 + ∆ w35 = − 1 . 2 − 0 . 0067 = − 1 . 2067 w 45 = w 45 + ∆ w 45 = 1 . 1 − 0 . 0112 = 1 .0888 θ 3 = θ 3 + ∆ θ 3 = 0 . 8 − 0 .0038 = 0 . 7962 θ 4 = θ 4 + ∆ θ 4 = − 0 . 1 + 0 . 0015 = − 0 .0985 θ 5 = θ 5 + ∆ θ 5 = 0 . 3 + 0 . 0127 = 0 . 3127  The training process is repeated until the sum of squared errors is less than 0.001. © Negnevitsky, Pearson Education, 2005 43
  • 44. Learning curve for operation Exclusive-OR 10 Sum-Squared Network Error for 224 Epochs 1 Sum-Squared Error 10 0 10 -1 10 -2 10 -3 10 -4 0 50 © Negnevitsky, Pearson Education, 2005 100 Epoch 150 200 44
  • 45. Final results of three-layer network learning Inputs x1 x2 1 0 1 0 1 1 0 0 Desired output Actual output yd y5 0.0155 0.9849 0.9849 0.0175 0 1 1 0 © Negnevitsky, Pearson Education, 2005 e Sum of squared errors 0.0010 45
  • 46. Network represented by McCulloch-Pitts model for solving the Exclusive-OR operation −1 +1.5 x1 1 +1.0 3 −1 −2.0 +1.0 +0.5 5 x2 2 +1.0 +1.0 y5 +1.0 4 +0.5 −1 © Negnevitsky, Pearson Education, 2005 46
  • 47. Decision boundaries x2 x2 x2 x1 + x 2 – 1.5 = 0 x 1 + x2 – 0.5 = 0 1 1 1 x1 x1 0 1 (a) 0 1 (b) x1 0 1 (c) (a) Decision boundary constructed by hidden neuron 3; (b) Decision boundary constructed by hidden neuron 4; (c) Decision boundaries constructed by the complete three-layer network © Negnevitsky, Pearson Education, 2005 47
  • 48. Accelerated learning in multilayer neural networks  A multilayer network learns much faster when the sigmoidal activation function is represented by a hyperbolic tangent: 2a tan h Y = −a 1 + e −bX where a and b are constants. Suitable values for a and b are: a = 1.716 and b = 0.667 © Negnevitsky, Pearson Education, 2005 48
  • 49.  We also can accelerate training by including a momentum term in the delta rule: ∆w jk ( p) = ×∆w jk ( p − 1) + ×y j ( p ) × k ( p ) where β is a positive number (0 ≤ β < 1) called the momentum constant. Typically, the momentum constant is set to 0.95. This equation is called the generalised delta rule. © Negnevitsky, Pearson Education, 2005 49
  • 50. Learning with momentum for operation Exclusive-OR 10 Training for 126 Epochs 2 10 1 10 0 10 -1 10 -2 10 -3 10 -4 0 20 40 60 Epoch 80 100 120 1.5 Learning Rate 1 0.5 0 -0.5 -1 0 20 40 60 80 100 120 140 Epoch © Negnevitsky, Pearson Education, 2005 50
  • 51. Learning with adaptive learning rate To accelerate the convergence and yet avoid the danger of instability, we can apply two heuristics: Heuristic 1 If the change of the sum of squared errors has the same algebraic sign for several consequent epochs, then the learning rate parameter, α, should be increased. Heuristic 2 If the algebraic sign of the change of the sum of squared errors alternates for several consequent epochs, then the learning rate parameter, α, should be decreased. © Negnevitsky, Pearson Education, 2005 51
  • 52.  Adapting the learning rate requires some changes in the back-propagation algorithm.  If the sum of squared errors at the current epoch exceeds the previous value by more than a predefined ratio (typically 1.04), the learning rate parameter is decreased (typically by multiplying by 0.7) and new weights and thresholds are calculated.  If the error is less than the previous one, the learning rate is increased (typically by multiplying by 1.05). © Negnevitsky, Pearson Education, 2005 52
  • 53. Learning with adaptive learning rate Sum-Squared Erro 10 Training for 103 Epochs 2 10 1 10 0 10 -1 10 -2 10 -3 10 -4 0 10 20 30 40 50 60 Epoch 70 80 90 100 1 Learning Rate 0. 8 0. 6 0. 4 0. 2 0 0 20 © Negnevitsky, Pearson Education, 2005 40 60 Epoch 80 100 120 53
  • 54. Learning with momentum and adaptive learning rate Sum-Squared Erro 10 Training for 85 Epochs 2 10 1 10 0 10 -1 10 -2 10 -3 10 -4 0 10 0 10 20 30 40 Epoch 50 60 70 80 Learning Rate 2.5 2 1.5 1 0.5 0 20 30 40 50 60 70 80 90 Epoch © Negnevitsky, Pearson Education, 2005 54
  • 55. The Hopfield Network  Neural networks were designed on analogy with the brain. The brain’s memory, however, works by association. For example, we can recognise a familiar face even in an unfamiliar environment within 100-200 ms. We can also recall a complete sensory experience, including sounds and scenes, when we hear only a few bars of music. The brain routinely associates one thing with another. © Negnevitsky, Pearson Education, 2005 55
  • 56.  Multilayer neural networks trained with the backpropagation algorithm are used for pattern recognition problems. However, to emulate the human memory’s associative characteristics we need a different type of network: a recurrent neural network.  A recurrent neural network has feedback loops from its outputs to its inputs. The presence of such loops has a profound impact on the learning capability of the network. © Negnevitsky, Pearson Education, 2005 56
  • 57.  The stability of recurrent networks intrigued several researchers in the 1960s and 1970s. However, none was able to predict which network would be stable, and some researchers were pessimistic about finding a solution at all. The problem was solved only in 1982, when John Hopfield formulated the physical principle of storing information in a dynamically stable network. © Negnevitsky, Pearson Education, 2005 57
  • 58. x1 1 y1 x2 2 y2 xi i yi xn n yn © Negnevitsky, Pearson Education, 2005 Output Signals Input Signals Single-layer n-neuron Hopfield network 58
  • 59.  The Hopfield network uses McCulloch and Pitts neurons with the sign activation function as its computing element: +1, if X > 0 sign  Y = −1, if X < 0  Y , if X = 0  © Negnevitsky, Pearson Education, 2005 59
  • 60.  The current state of the Hopfield network is determined by the current outputs of all neurons, y1, y2, . . ., yn. Thus, for a single-layer n-neuron network, the state can be defined by the state vector as:  y1   y   2  Y =      yn    © Negnevitsky, Pearson Education, 2005 60
  • 61.  In the Hopfield network, synaptic weights between neurons are usually represented in matrix form as follows: W= M T YmYm − M I ∑ m=1 where M is the number of states to be memorised by the network, Ym is the n-dimensional binary vector, I is n × n identity matrix, and superscript T denotes matrix transposition. © Negnevitsky, Pearson Education, 2005 61
  • 62. Possible states for the three-neuron Hopfield network y2 (−1,1, −1) (1, 1, −1) (1, 1, 1) (−1, 1, 1) y1 0 (1,−1,−1) (−1,−1,−1) y3 (−1,−1, 1) © Negnevitsky, Pearson Education, 2005 (1,−1, 1) 62
  • 63.  The stable state-vertex is determined by the weight matrix W, the current input vector X, and the threshold matrix θ . If the input vector is partially incorrect or incomplete, the initial state will converge into the stable state-vertex after a few iterations.  Suppose, for instance, that our network is required to memorise two opposite states, (1, 1, 1) and (−1, −1, −1). Thus, 1 Y1 = 1  1  − 1 Y2 = − 1   − 1   T or Y1 = 1 1 1 T Y2 = − 1 − 1 − 1 where Y1 and Y2 are the three-dimensional vectors. © Negnevitsky, Pearson Education, 2005 63
  • 64.  The 3 × 3 identity matrix I is 1 0 0 I = 0 1 0   0 0 1    Thus, we can now determine the weight matrix as follows: 1 −1 1 0 0 0 W = 1 1 1 1 + −1 −1 −1 −1 − 2 0 1 0 = 2       1 −1 0 0 1 2       2 0 2 2 2  0   Next, the network is tested by the sequence of input vectors, X1 and X2, which are equal to the output (or target) vectors Y1 and Y2, respectively. © Negnevitsky, Pearson Education, 2005 64
  • 65.  First, we activate the Hopfield network by applying the input vector X. Then, we calculate the actual output vector Y, and finally, we compare the result with the initial input vector X. 0  Y1 = sign 2 2  2 0 2 2 1 0  1  1 − 0  = 1 2        0 1 0  1       0  Y2 = sign 2 2  2 0 2 2 −1 0 −1  −1 − 0 = −1 2       0 −1 0 −1        © Negnevitsky, Pearson Education, 2005 65
  • 66.  The remaining six states are all unstable. However, stable states (also called fundamental memories) are capable of attracting states that are close to them.  The fundamental memory (1, 1, 1) attracts unstable states (−1, 1, 1), (1, −1, 1) and (1, 1, −1). Each of these unstable states represents a single error, compared to the fundamental memory (1, 1, 1).  The fundamental memory (−1, −1, −1) attracts unstable states (−1, −1, 1), (−1, 1, −1) and (1, −1, −1).  Thus, the Hopfield network can act as an error correction network. © Negnevitsky, Pearson Education, 2005 66
  • 67. Storage capacity of the Hopfield network  Storage capacity is or the largest number of fundamental memories that can be stored and retrieved correctly.  The maximum number of fundamental memories Mmax that can be stored in the n-neuron recurrent network is limited by M max = 0.15n © Negnevitsky, Pearson Education, 2005 67
  • 68. Bidirectional associative memory (BAM)  The Hopfield network represents an autoassociative type of memory − it can retrieve a corrupted or incomplete memory but cannot associate this memory with another different memory.  Human memory is essentially associative. One thing may remind us of another, and that of another, and so on. We use a chain of mental associations to recover a lost memory. If we forget where we left an umbrella, we try to recall where we last had it, what we were doing, and who we were talking to. We attempt to establish a chain of associations, and thereby to restore a lost memory. © Negnevitsky, Pearson Education, 2005 68
  • 69.  To associate one memory with another, we need a recurrent neural network capable of accepting an input pattern on one set of neurons and producing a related, but different, output pattern on another set of neurons.  Bidirectional associative memory (BAM), first proposed by Bart Kosko, is a heteroassociative network. It associates patterns from one set, set A, to patterns from another set, set B, and vice versa. Like a Hopfield network, the BAM can generalise and also produce correct outputs despite corrupted or incomplete inputs. © Negnevitsky, Pearson Education, 2005 69
  • 70. BAM operation x1(p) x1(p+1) 1 1 x2 (p) 2 xi (p) i y1(p) 2 y2(p) j yj(p) m xn(p) 1 2 xi(p+1) i xn(p+1) Output layer (a) Forward direction. © Negnevitsky, Pearson Education, 2005 y1(p) 2 y2(p) j yj(p) m x2(p+1) ym(p) n Input layer 1 ym(p) n Input layer Output layer (b) Backward direction. 70
  • 71. The basic idea behind the BAM is to store pattern pairs so that when n-dimensional vector X from set A is presented as input, the BAM recalls m-dimensional vector Y from set B, but when Y is presented as input, the BAM recalls X. © Negnevitsky, Pearson Education, 2005 71
  • 72.  To develop the BAM, we need to create a correlation matrix for each pattern pair we want to store. The correlation matrix is the matrix product of the input vector X, and the transpose of the output vector YT. The BAM weight matrix is the sum of all correlation matrices, that is, W= M T Xm Ym ∑ m=1 where M is the number of pattern pairs to be stored in the BAM. © Negnevitsky, Pearson Education, 2005 72
  • 73. Stability and storage capacity of the BAM  The BAM is unconditionally stable. This means that any set of associations can be learned without risk of instability.  The maximum number of associations to be stored in the BAM should not exceed the number of neurons in the smaller layer.  The more serious problem with the BAM is incorrect convergence. The BAM may not always produce the closest association. In fact, a stable association may be only slightly related to the initial input vector. © Negnevitsky, Pearson Education, 2005 73

Notas do Editor

  1. 44